4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
28 * Pool import support functions.
30 * To import a pool, we rely on reading the configuration information from the
31 * ZFS label of each device. If we successfully read the label, then we
32 * organize the configuration information in the following hierarchy:
34 * pool guid -> toplevel vdev guid -> label txg
36 * Duplicate entries matching this same tuple will be discarded. Once we have
37 * examined every device, we pick the best label txg config for each toplevel
38 * vdev. We then arrange these toplevel vdevs into a complete pool config, and
39 * update any paths that have changed. Finally, we attempt to import the pool
40 * using our derived config, and record the results.
54 #include <thread_pool.h>
57 #include <sys/vdev_impl.h>
60 #include "libzfs_impl.h"
63 * Intermediate structures used to gather configuration information.
65 typedef struct config_entry {
68 struct config_entry *ce_next;
71 typedef struct vdev_entry {
73 config_entry_t *ve_configs;
74 struct vdev_entry *ve_next;
77 typedef struct pool_entry {
79 vdev_entry_t *pe_vdevs;
80 struct pool_entry *pe_next;
83 typedef struct name_entry {
86 struct name_entry *ne_next;
89 typedef struct pool_list {
95 get_devid(const char *path)
102 if ((fd = open(path, O_RDONLY)) < 0)
107 if (devid_get(fd, &devid) == 0) {
108 if (devid_get_minor_name(fd, &minor) == 0)
109 ret = devid_str_encode(devid, minor);
111 devid_str_free(minor);
124 * Go through and fix up any path and/or devid information for the given vdev
128 fix_paths(nvlist_t *nv, name_entry_t *names)
133 name_entry_t *ne, *best;
137 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
138 &child, &children) == 0) {
139 for (c = 0; c < children; c++)
140 if (fix_paths(child[c], names) != 0)
146 * This is a leaf (file or disk) vdev. In either case, go through
147 * the name list and see if we find a matching guid. If so, replace
148 * the path and see if we can calculate a new devid.
150 * There may be multiple names associated with a particular guid, in
151 * which case we have overlapping slices or multiple paths to the same
152 * disk. If this is the case, then we want to pick the path that is
153 * the most similar to the original, where "most similar" is the number
154 * of matching characters starting from the end of the path. This will
155 * preserve slice numbers even if the disks have been reorganized, and
156 * will also catch preferred disk names if multiple paths exist.
158 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0);
159 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
164 for (ne = names; ne != NULL; ne = ne->ne_next) {
165 if (ne->ne_guid == guid) {
166 const char *src, *dst;
174 src = ne->ne_name + strlen(ne->ne_name) - 1;
175 dst = path + strlen(path) - 1;
176 for (count = 0; src >= ne->ne_name && dst >= path;
177 src--, dst--, count++)
182 * At this point, 'count' is the number of characters
183 * matched from the end.
185 if (count > matched || best == NULL) {
195 if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
198 if ((devid = get_devid(best->ne_name)) == NULL) {
199 (void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
201 if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0)
203 devid_str_free(devid);
210 * Add the given configuration to the list of known devices.
213 add_config(libzfs_handle_t *hdl, pool_list_t *pl, const char *path,
216 uint64_t pool_guid, vdev_guid, top_guid, txg, state;
223 * If this is a hot spare not currently in use or level 2 cache
224 * device, add it to the list of names to translate, but don't do
227 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
229 (state == POOL_STATE_SPARE || state == POOL_STATE_L2CACHE) &&
230 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid) == 0) {
231 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
234 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
238 ne->ne_guid = vdev_guid;
239 ne->ne_next = pl->names;
245 * If we have a valid config but cannot read any of these fields, then
246 * it means we have a half-initialized label. In vdev_label_init()
247 * we write a label with txg == 0 so that we can identify the device
248 * in case the user refers to the same disk later on. If we fail to
249 * create the pool, we'll be left with a label in this state
250 * which should not be considered part of a valid pool.
252 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
254 nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
256 nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID,
258 nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
259 &txg) != 0 || txg == 0) {
265 * First, see if we know about this pool. If not, then add it to the
266 * list of known pools.
268 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
269 if (pe->pe_guid == pool_guid)
274 if ((pe = zfs_alloc(hdl, sizeof (pool_entry_t))) == NULL) {
278 pe->pe_guid = pool_guid;
279 pe->pe_next = pl->pools;
284 * Second, see if we know about this toplevel vdev. Add it if its
287 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
288 if (ve->ve_guid == top_guid)
293 if ((ve = zfs_alloc(hdl, sizeof (vdev_entry_t))) == NULL) {
297 ve->ve_guid = top_guid;
298 ve->ve_next = pe->pe_vdevs;
303 * Third, see if we have a config with a matching transaction group. If
304 * so, then we do nothing. Otherwise, add it to the list of known
307 for (ce = ve->ve_configs; ce != NULL; ce = ce->ce_next) {
308 if (ce->ce_txg == txg)
313 if ((ce = zfs_alloc(hdl, sizeof (config_entry_t))) == NULL) {
318 ce->ce_config = config;
319 ce->ce_next = ve->ve_configs;
326 * At this point we've successfully added our config to the list of
327 * known configs. The last thing to do is add the vdev guid -> path
328 * mappings so that we can fix up the configuration as necessary before
331 if ((ne = zfs_alloc(hdl, sizeof (name_entry_t))) == NULL)
334 if ((ne->ne_name = zfs_strdup(hdl, path)) == NULL) {
339 ne->ne_guid = vdev_guid;
340 ne->ne_next = pl->names;
347 * Returns true if the named pool matches the given GUID.
350 pool_active(libzfs_handle_t *hdl, const char *name, uint64_t guid,
356 if (zpool_open_silent(hdl, name, &zhp) != 0)
364 verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_POOL_GUID,
369 *isactive = (theguid == guid);
374 refresh_config(libzfs_handle_t *hdl, nvlist_t *config)
377 zfs_cmd_t zc = { 0 };
380 if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0)
383 if (zcmd_alloc_dst_nvlist(hdl, &zc,
384 zc.zc_nvlist_conf_size * 2) != 0) {
385 zcmd_free_nvlists(&zc);
389 while ((err = ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_TRYIMPORT,
390 &zc)) != 0 && errno == ENOMEM) {
391 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
392 zcmd_free_nvlists(&zc);
398 zcmd_free_nvlists(&zc);
402 if (zcmd_read_dst_nvlist(hdl, &zc, &nvl) != 0) {
403 zcmd_free_nvlists(&zc);
407 zcmd_free_nvlists(&zc);
412 * Determine if the vdev id is a hole in the namespace.
415 vdev_is_hole(uint64_t *hole_array, uint_t holes, uint_t id)
417 for (int c = 0; c < holes; c++) {
419 /* Top-level is a hole */
420 if (hole_array[c] == id)
427 * Convert our list of pools into the definitive set of configurations. We
428 * start by picking the best config for each toplevel vdev. Once that's done,
429 * we assemble the toplevel vdevs into a full config for the pool. We make a
430 * pass to fix up any incorrect paths, and then add it to the main list to
431 * return to the user.
434 get_configs(libzfs_handle_t *hdl, pool_list_t *pl, boolean_t active_ok)
439 nvlist_t *ret = NULL, *config = NULL, *tmp, *nvtop, *nvroot;
440 nvlist_t **spares, **l2cache;
441 uint_t i, nspares, nl2cache;
442 boolean_t config_seen;
444 char *name, *hostname;
447 nvlist_t **child = NULL;
449 uint64_t *hole_array, max_id;
454 boolean_t found_one = B_FALSE;
455 boolean_t valid_top_config = B_FALSE;
457 if (nvlist_alloc(&ret, 0, 0) != 0)
460 for (pe = pl->pools; pe != NULL; pe = pe->pe_next) {
461 uint64_t id, max_txg = 0;
463 if (nvlist_alloc(&config, NV_UNIQUE_NAME, 0) != 0)
465 config_seen = B_FALSE;
468 * Iterate over all toplevel vdevs. Grab the pool configuration
469 * from the first one we find, and then go through the rest and
470 * add them as necessary to the 'vdevs' member of the config.
472 for (ve = pe->pe_vdevs; ve != NULL; ve = ve->ve_next) {
475 * Determine the best configuration for this vdev by
476 * selecting the config with the latest transaction
480 for (ce = ve->ve_configs; ce != NULL;
483 if (ce->ce_txg > best_txg) {
485 best_txg = ce->ce_txg;
490 * We rely on the fact that the max txg for the
491 * pool will contain the most up-to-date information
492 * about the valid top-levels in the vdev namespace.
494 if (best_txg > max_txg) {
495 (void) nvlist_remove(config,
496 ZPOOL_CONFIG_VDEV_CHILDREN,
498 (void) nvlist_remove(config,
499 ZPOOL_CONFIG_HOLE_ARRAY,
500 DATA_TYPE_UINT64_ARRAY);
506 valid_top_config = B_FALSE;
508 if (nvlist_lookup_uint64(tmp,
509 ZPOOL_CONFIG_VDEV_CHILDREN, &max_id) == 0) {
510 verify(nvlist_add_uint64(config,
511 ZPOOL_CONFIG_VDEV_CHILDREN,
513 valid_top_config = B_TRUE;
516 if (nvlist_lookup_uint64_array(tmp,
517 ZPOOL_CONFIG_HOLE_ARRAY, &hole_array,
519 verify(nvlist_add_uint64_array(config,
520 ZPOOL_CONFIG_HOLE_ARRAY,
521 hole_array, holes) == 0);
527 * Copy the relevant pieces of data to the pool
533 * comment (if available)
535 * hostid (if available)
536 * hostname (if available)
538 uint64_t state, version;
539 char *comment = NULL;
541 version = fnvlist_lookup_uint64(tmp,
542 ZPOOL_CONFIG_VERSION);
543 fnvlist_add_uint64(config,
544 ZPOOL_CONFIG_VERSION, version);
545 guid = fnvlist_lookup_uint64(tmp,
546 ZPOOL_CONFIG_POOL_GUID);
547 fnvlist_add_uint64(config,
548 ZPOOL_CONFIG_POOL_GUID, guid);
549 name = fnvlist_lookup_string(tmp,
550 ZPOOL_CONFIG_POOL_NAME);
551 fnvlist_add_string(config,
552 ZPOOL_CONFIG_POOL_NAME, name);
554 if (nvlist_lookup_string(tmp,
555 ZPOOL_CONFIG_COMMENT, &comment) == 0)
556 fnvlist_add_string(config,
557 ZPOOL_CONFIG_COMMENT, comment);
559 state = fnvlist_lookup_uint64(tmp,
560 ZPOOL_CONFIG_POOL_STATE);
561 fnvlist_add_uint64(config,
562 ZPOOL_CONFIG_POOL_STATE, state);
565 if (nvlist_lookup_uint64(tmp,
566 ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
567 fnvlist_add_uint64(config,
568 ZPOOL_CONFIG_HOSTID, hostid);
569 hostname = fnvlist_lookup_string(tmp,
570 ZPOOL_CONFIG_HOSTNAME);
571 fnvlist_add_string(config,
572 ZPOOL_CONFIG_HOSTNAME, hostname);
575 config_seen = B_TRUE;
579 * Add this top-level vdev to the child array.
581 verify(nvlist_lookup_nvlist(tmp,
582 ZPOOL_CONFIG_VDEV_TREE, &nvtop) == 0);
583 verify(nvlist_lookup_uint64(nvtop, ZPOOL_CONFIG_ID,
586 if (id >= children) {
589 newchild = zfs_alloc(hdl, (id + 1) *
590 sizeof (nvlist_t *));
591 if (newchild == NULL)
594 for (c = 0; c < children; c++)
595 newchild[c] = child[c];
601 if (nvlist_dup(nvtop, &child[id], 0) != 0)
607 * If we have information about all the top-levels then
608 * clean up the nvlist which we've constructed. This
609 * means removing any extraneous devices that are
610 * beyond the valid range or adding devices to the end
611 * of our array which appear to be missing.
613 if (valid_top_config) {
614 if (max_id < children) {
615 for (c = max_id; c < children; c++)
616 nvlist_free(child[c]);
618 } else if (max_id > children) {
621 newchild = zfs_alloc(hdl, (max_id) *
622 sizeof (nvlist_t *));
623 if (newchild == NULL)
626 for (c = 0; c < children; c++)
627 newchild[c] = child[c];
635 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
639 * The vdev namespace may contain holes as a result of
640 * device removal. We must add them back into the vdev
641 * tree before we process any missing devices.
644 ASSERT(valid_top_config);
646 for (c = 0; c < children; c++) {
649 if (child[c] != NULL ||
650 !vdev_is_hole(hole_array, holes, c))
653 if (nvlist_alloc(&holey, NV_UNIQUE_NAME,
658 * Holes in the namespace are treated as
659 * "hole" top-level vdevs and have a
660 * special flag set on them.
662 if (nvlist_add_string(holey,
664 VDEV_TYPE_HOLE) != 0 ||
665 nvlist_add_uint64(holey,
666 ZPOOL_CONFIG_ID, c) != 0 ||
667 nvlist_add_uint64(holey,
668 ZPOOL_CONFIG_GUID, 0ULL) != 0)
675 * Look for any missing top-level vdevs. If this is the case,
676 * create a faked up 'missing' vdev as a placeholder. We cannot
677 * simply compress the child array, because the kernel performs
678 * certain checks to make sure the vdev IDs match their location
679 * in the configuration.
681 for (c = 0; c < children; c++) {
682 if (child[c] == NULL) {
684 if (nvlist_alloc(&missing, NV_UNIQUE_NAME,
687 if (nvlist_add_string(missing,
689 VDEV_TYPE_MISSING) != 0 ||
690 nvlist_add_uint64(missing,
691 ZPOOL_CONFIG_ID, c) != 0 ||
692 nvlist_add_uint64(missing,
693 ZPOOL_CONFIG_GUID, 0ULL) != 0) {
694 nvlist_free(missing);
702 * Put all of this pool's top-level vdevs into a root vdev.
704 if (nvlist_alloc(&nvroot, NV_UNIQUE_NAME, 0) != 0)
706 if (nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
707 VDEV_TYPE_ROOT) != 0 ||
708 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) != 0 ||
709 nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, guid) != 0 ||
710 nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
711 child, children) != 0) {
716 for (c = 0; c < children; c++)
717 nvlist_free(child[c]);
723 * Go through and fix up any paths and/or devids based on our
724 * known list of vdev GUID -> path mappings.
726 if (fix_paths(nvroot, pl->names) != 0) {
732 * Add the root vdev to this pool's configuration.
734 if (nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
742 * zdb uses this path to report on active pools that were
743 * imported or created using -R.
749 * Determine if this pool is currently active, in which case we
750 * can't actually import it.
752 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
754 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
757 if (pool_active(hdl, name, guid, &isactive) != 0)
766 if ((nvl = refresh_config(hdl, config)) == NULL) {
776 * Go through and update the paths for spares, now that we have
779 verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
781 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
782 &spares, &nspares) == 0) {
783 for (i = 0; i < nspares; i++) {
784 if (fix_paths(spares[i], pl->names) != 0)
790 * Update the paths for l2cache devices.
792 if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
793 &l2cache, &nl2cache) == 0) {
794 for (i = 0; i < nl2cache; i++) {
795 if (fix_paths(l2cache[i], pl->names) != 0)
801 * Restore the original information read from the actual label.
803 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTID,
805 (void) nvlist_remove(config, ZPOOL_CONFIG_HOSTNAME,
808 verify(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
810 verify(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
816 * Add this pool to the list of configs.
818 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
820 if (nvlist_add_nvlist(ret, name, config) != 0)
836 (void) no_memory(hdl);
840 for (c = 0; c < children; c++)
841 nvlist_free(child[c]);
848 * Return the offset of the given label.
851 label_offset(uint64_t size, int l)
853 ASSERT(P2PHASE_TYPED(size, sizeof (vdev_label_t), uint64_t) == 0);
854 return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
855 0 : size - VDEV_LABELS * sizeof (vdev_label_t)));
859 * Given a file descriptor, read the label information and return an nvlist
860 * describing the configuration, if there is one.
863 zpool_read_label(int fd, nvlist_t **config)
865 struct stat64 statbuf;
868 uint64_t state, txg, size;
872 if (fstat64(fd, &statbuf) == -1)
874 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
876 if ((label = malloc(sizeof (vdev_label_t))) == NULL)
879 for (l = 0; l < VDEV_LABELS; l++) {
880 if (pread64(fd, label, sizeof (vdev_label_t),
881 label_offset(size, l)) != sizeof (vdev_label_t))
884 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
885 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0)
888 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
889 &state) != 0 || state > POOL_STATE_L2CACHE) {
890 nvlist_free(*config);
894 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
895 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
896 &txg) != 0 || txg == 0)) {
897 nvlist_free(*config);
910 typedef struct rdsk_node {
913 libzfs_handle_t *rn_hdl;
917 boolean_t rn_nozpool;
921 slice_cache_compare(const void *arg1, const void *arg2)
923 const char *nm1 = ((rdsk_node_t *)arg1)->rn_name;
924 const char *nm2 = ((rdsk_node_t *)arg2)->rn_name;
925 char *nm1slice, *nm2slice;
929 * slices zero and two are the most likely to provide results,
932 nm1slice = strstr(nm1, "s0");
933 nm2slice = strstr(nm2, "s0");
934 if (nm1slice && !nm2slice) {
937 if (!nm1slice && nm2slice) {
940 nm1slice = strstr(nm1, "s2");
941 nm2slice = strstr(nm2, "s2");
942 if (nm1slice && !nm2slice) {
945 if (!nm1slice && nm2slice) {
949 rv = strcmp(nm1, nm2);
952 return (rv > 0 ? 1 : -1);
957 check_one_slice(avl_tree_t *r, char *diskname, uint_t partno,
958 diskaddr_t size, uint_t blksz)
962 char sname[MAXNAMELEN];
964 tmpnode.rn_name = &sname[0];
965 (void) snprintf(tmpnode.rn_name, MAXNAMELEN, "%s%u",
968 * protect against division by zero for disk labels that
969 * contain a bogus sector size
973 /* too small to contain a zpool? */
974 if ((size < (SPA_MINDEVSIZE / blksz)) &&
975 (node = avl_find(r, &tmpnode, NULL)))
976 node->rn_nozpool = B_TRUE;
981 nozpool_all_slices(avl_tree_t *r, const char *sname)
984 char diskname[MAXNAMELEN];
988 (void) strncpy(diskname, sname, MAXNAMELEN);
989 if (((ptr = strrchr(diskname, 's')) == NULL) &&
990 ((ptr = strrchr(diskname, 'p')) == NULL))
994 for (i = 0; i < NDKMAP; i++)
995 check_one_slice(r, diskname, i, 0, 1);
997 for (i = 0; i <= FD_NUMPART; i++)
998 check_one_slice(r, diskname, i, 0, 1);
1004 check_slices(avl_tree_t *r, int fd, const char *sname)
1006 struct extvtoc vtoc;
1008 char diskname[MAXNAMELEN];
1012 (void) strncpy(diskname, sname, MAXNAMELEN);
1013 if ((ptr = strrchr(diskname, 's')) == NULL || !isdigit(ptr[1]))
1017 if (read_extvtoc(fd, &vtoc) >= 0) {
1018 for (i = 0; i < NDKMAP; i++)
1019 check_one_slice(r, diskname, i,
1020 vtoc.v_part[i].p_size, vtoc.v_sectorsz);
1021 } else if (efi_alloc_and_read(fd, &gpt) >= 0) {
1023 * on x86 we'll still have leftover links that point
1024 * to slices s[9-15], so use NDKMAP instead
1026 for (i = 0; i < NDKMAP; i++)
1027 check_one_slice(r, diskname, i,
1028 gpt->efi_parts[i].p_size, gpt->efi_lbasize);
1029 /* nodes p[1-4] are never used with EFI labels */
1031 for (i = 1; i <= FD_NUMPART; i++)
1032 check_one_slice(r, diskname, i, 0, 1);
1039 zpool_open_func(void *arg)
1041 rdsk_node_t *rn = arg;
1042 struct stat64 statbuf;
1048 if ((fd = openat64(rn->rn_dfd, rn->rn_name, O_RDONLY)) < 0) {
1049 /* symlink to a device that's no longer there */
1050 if (errno == ENOENT)
1051 nozpool_all_slices(rn->rn_avl, rn->rn_name);
1055 * Ignore failed stats. We only want regular
1056 * files, character devs and block devs.
1058 if (fstat64(fd, &statbuf) != 0 ||
1059 (!S_ISREG(statbuf.st_mode) &&
1060 !S_ISCHR(statbuf.st_mode) &&
1061 !S_ISBLK(statbuf.st_mode))) {
1065 /* this file is too small to hold a zpool */
1067 if (S_ISREG(statbuf.st_mode) &&
1068 statbuf.st_size < SPA_MINDEVSIZE) {
1071 } else if (!S_ISREG(statbuf.st_mode)) {
1073 * Try to read the disk label first so we don't have to
1074 * open a bunch of minor nodes that can't have a zpool.
1076 check_slices(rn->rn_avl, fd, rn->rn_name);
1079 if (statbuf.st_size < SPA_MINDEVSIZE) {
1085 if ((zpool_read_label(fd, &config)) != 0) {
1087 (void) no_memory(rn->rn_hdl);
1093 rn->rn_config = config;
1094 if (config != NULL) {
1095 assert(rn->rn_nozpool == B_FALSE);
1100 * Given a file descriptor, clear (zero) the label information. This function
1101 * is used in the appliance stack as part of the ZFS sysevent module and
1102 * to implement the "zpool labelclear" command.
1105 zpool_clear_label(int fd)
1107 struct stat64 statbuf;
1109 vdev_label_t *label;
1112 if (fstat64(fd, &statbuf) == -1)
1114 size = P2ALIGN_TYPED(statbuf.st_size, sizeof (vdev_label_t), uint64_t);
1116 if ((label = calloc(sizeof (vdev_label_t), 1)) == NULL)
1119 for (l = 0; l < VDEV_LABELS; l++) {
1120 if (pwrite64(fd, label, sizeof (vdev_label_t),
1121 label_offset(size, l)) != sizeof (vdev_label_t))
1130 * Given a list of directories to search, find all pools stored on disk. This
1131 * includes partial pools which are not available to import. If no args are
1132 * given (argc is 0), then the default directory (/dev/dsk) is searched.
1133 * poolname or guid (but not both) are provided by the caller when trying
1134 * to import a specific pool.
1137 zpool_find_import_impl(libzfs_handle_t *hdl, importargs_t *iarg)
1139 int i, dirs = iarg->paths;
1141 struct dirent64 *dp;
1142 char path[MAXPATHLEN];
1143 char *end, **dir = iarg->path;
1145 nvlist_t *ret = NULL;
1146 static char *default_dir = "/dev";
1147 pool_list_t pools = { 0 };
1148 pool_entry_t *pe, *penext;
1149 vdev_entry_t *ve, *venext;
1150 config_entry_t *ce, *cenext;
1151 name_entry_t *ne, *nenext;
1152 avl_tree_t slice_cache;
1162 * Go through and read the label configuration information from every
1163 * possible device, organizing the information according to pool GUID
1164 * and toplevel GUID.
1166 for (i = 0; i < dirs; i++) {
1171 /* use realpath to normalize the path */
1172 if (realpath(dir[i], path) == 0) {
1173 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1174 dgettext(TEXT_DOMAIN, "cannot open '%s'"), dir[i]);
1177 end = &path[strlen(path)];
1180 pathleft = &path[sizeof (path)] - end;
1183 * Using raw devices instead of block devices when we're
1184 * reading the labels skips a bunch of slow operations during
1185 * close(2) processing, so we replace /dev/dsk with /dev/rdsk.
1187 if (strcmp(path, "/dev/dsk/") == 0)
1192 if ((dfd = open64(rdsk, O_RDONLY)) < 0 ||
1193 (dirp = fdopendir(dfd)) == NULL) {
1194 zfs_error_aux(hdl, strerror(errno));
1195 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1196 dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1201 avl_create(&slice_cache, slice_cache_compare,
1202 sizeof (rdsk_node_t), offsetof(rdsk_node_t, rn_node));
1204 if (strcmp(rdsk, "/dev/") == 0) {
1208 struct gprovider *pp;
1210 errno = geom_gettree(&mesh);
1212 zfs_error_aux(hdl, strerror(errno));
1213 (void) zfs_error_fmt(hdl, EZFS_BADPATH,
1214 dgettext(TEXT_DOMAIN, "cannot get GEOM tree"));
1218 LIST_FOREACH(mp, &mesh.lg_class, lg_class) {
1219 LIST_FOREACH(gp, &mp->lg_geom, lg_geom) {
1220 LIST_FOREACH(pp, &gp->lg_provider, lg_provider) {
1221 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1222 slice->rn_name = zfs_strdup(hdl, pp->lg_name);
1223 slice->rn_avl = &slice_cache;
1224 slice->rn_dfd = dfd;
1225 slice->rn_hdl = hdl;
1226 slice->rn_nozpool = B_FALSE;
1227 avl_add(&slice_cache, slice);
1232 geom_deletetree(&mesh);
1237 * This is not MT-safe, but we have no MT consumers of libzfs
1239 while ((dp = readdir64(dirp)) != NULL) {
1240 const char *name = dp->d_name;
1241 if (name[0] == '.' &&
1242 (name[1] == 0 || (name[1] == '.' && name[2] == 0)))
1245 slice = zfs_alloc(hdl, sizeof (rdsk_node_t));
1246 slice->rn_name = zfs_strdup(hdl, name);
1247 slice->rn_avl = &slice_cache;
1248 slice->rn_dfd = dfd;
1249 slice->rn_hdl = hdl;
1250 slice->rn_nozpool = B_FALSE;
1251 avl_add(&slice_cache, slice);
1255 * create a thread pool to do all of this in parallel;
1256 * rn_nozpool is not protected, so this is racy in that
1257 * multiple tasks could decide that the same slice can
1258 * not hold a zpool, which is benign. Also choose
1259 * double the number of processors; we hold a lot of
1260 * locks in the kernel, so going beyond this doesn't
1263 t = tpool_create(1, 2 * sysconf(_SC_NPROCESSORS_ONLN),
1265 for (slice = avl_first(&slice_cache); slice;
1266 (slice = avl_walk(&slice_cache, slice,
1268 (void) tpool_dispatch(t, zpool_open_func, slice);
1273 while ((slice = avl_destroy_nodes(&slice_cache,
1274 &cookie)) != NULL) {
1275 if (slice->rn_config != NULL) {
1276 nvlist_t *config = slice->rn_config;
1277 boolean_t matched = B_TRUE;
1279 if (iarg->poolname != NULL) {
1282 matched = nvlist_lookup_string(config,
1283 ZPOOL_CONFIG_POOL_NAME,
1285 strcmp(iarg->poolname, pname) == 0;
1286 } else if (iarg->guid != 0) {
1289 matched = nvlist_lookup_uint64(config,
1290 ZPOOL_CONFIG_POOL_GUID,
1292 iarg->guid == this_guid;
1295 nvlist_free(config);
1299 /* use the non-raw path for the config */
1300 (void) strlcpy(end, slice->rn_name, pathleft);
1301 if (add_config(hdl, &pools, path, config) != 0)
1304 free(slice->rn_name);
1307 avl_destroy(&slice_cache);
1309 (void) closedir(dirp);
1313 ret = get_configs(hdl, &pools, iarg->can_be_active);
1316 for (pe = pools.pools; pe != NULL; pe = penext) {
1317 penext = pe->pe_next;
1318 for (ve = pe->pe_vdevs; ve != NULL; ve = venext) {
1319 venext = ve->ve_next;
1320 for (ce = ve->ve_configs; ce != NULL; ce = cenext) {
1321 cenext = ce->ce_next;
1323 nvlist_free(ce->ce_config);
1331 for (ne = pools.names; ne != NULL; ne = nenext) {
1332 nenext = ne->ne_next;
1339 (void) closedir(dirp);
1345 zpool_find_import(libzfs_handle_t *hdl, int argc, char **argv)
1347 importargs_t iarg = { 0 };
1352 return (zpool_find_import_impl(hdl, &iarg));
1356 * Given a cache file, return the contents as a list of importable pools.
1357 * poolname or guid (but not both) are provided by the caller when trying
1358 * to import a specific pool.
1361 zpool_find_import_cached(libzfs_handle_t *hdl, const char *cachefile,
1362 char *poolname, uint64_t guid)
1366 struct stat64 statbuf;
1367 nvlist_t *raw, *src, *dst;
1374 verify(poolname == NULL || guid == 0);
1376 if ((fd = open(cachefile, O_RDONLY)) < 0) {
1377 zfs_error_aux(hdl, "%s", strerror(errno));
1378 (void) zfs_error(hdl, EZFS_BADCACHE,
1379 dgettext(TEXT_DOMAIN, "failed to open cache file"));
1383 if (fstat64(fd, &statbuf) != 0) {
1384 zfs_error_aux(hdl, "%s", strerror(errno));
1386 (void) zfs_error(hdl, EZFS_BADCACHE,
1387 dgettext(TEXT_DOMAIN, "failed to get size of cache file"));
1391 if ((buf = zfs_alloc(hdl, statbuf.st_size)) == NULL) {
1396 if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
1399 (void) zfs_error(hdl, EZFS_BADCACHE,
1400 dgettext(TEXT_DOMAIN,
1401 "failed to read cache file contents"));
1407 if (nvlist_unpack(buf, statbuf.st_size, &raw, 0) != 0) {
1409 (void) zfs_error(hdl, EZFS_BADCACHE,
1410 dgettext(TEXT_DOMAIN,
1411 "invalid or corrupt cache file contents"));
1418 * Go through and get the current state of the pools and refresh their
1421 if (nvlist_alloc(&pools, 0, 0) != 0) {
1422 (void) no_memory(hdl);
1428 while ((elem = nvlist_next_nvpair(raw, elem)) != NULL) {
1429 verify(nvpair_value_nvlist(elem, &src) == 0);
1431 verify(nvlist_lookup_string(src, ZPOOL_CONFIG_POOL_NAME,
1433 if (poolname != NULL && strcmp(poolname, name) != 0)
1436 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1439 verify(nvlist_lookup_uint64(src, ZPOOL_CONFIG_POOL_GUID,
1441 if (guid != this_guid)
1445 if (pool_active(hdl, name, this_guid, &active) != 0) {
1454 if ((dst = refresh_config(hdl, src)) == NULL) {
1460 if (nvlist_add_nvlist(pools, nvpair_name(elem), dst) != 0) {
1461 (void) no_memory(hdl);
1475 name_or_guid_exists(zpool_handle_t *zhp, void *data)
1477 importargs_t *import = data;
1480 if (import->poolname != NULL) {
1483 verify(nvlist_lookup_string(zhp->zpool_config,
1484 ZPOOL_CONFIG_POOL_NAME, &pool_name) == 0);
1485 if (strcmp(pool_name, import->poolname) == 0)
1490 verify(nvlist_lookup_uint64(zhp->zpool_config,
1491 ZPOOL_CONFIG_POOL_GUID, &pool_guid) == 0);
1492 if (pool_guid == import->guid)
1501 zpool_search_import(libzfs_handle_t *hdl, importargs_t *import)
1503 verify(import->poolname == NULL || import->guid == 0);
1506 import->exists = zpool_iter(hdl, name_or_guid_exists, import);
1508 if (import->cachefile != NULL)
1509 return (zpool_find_import_cached(hdl, import->cachefile,
1510 import->poolname, import->guid));
1512 return (zpool_find_import_impl(hdl, import));
1516 find_guid(nvlist_t *nv, uint64_t guid)
1522 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &tmp) == 0);
1526 if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1527 &child, &children) == 0) {
1528 for (c = 0; c < children; c++)
1529 if (find_guid(child[c], guid))
1536 typedef struct aux_cbdata {
1537 const char *cb_type;
1539 zpool_handle_t *cb_zhp;
1543 find_aux(zpool_handle_t *zhp, void *data)
1545 aux_cbdata_t *cbp = data;
1551 verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1554 if (nvlist_lookup_nvlist_array(nvroot, cbp->cb_type,
1555 &list, &count) == 0) {
1556 for (i = 0; i < count; i++) {
1557 verify(nvlist_lookup_uint64(list[i],
1558 ZPOOL_CONFIG_GUID, &guid) == 0);
1559 if (guid == cbp->cb_guid) {
1571 * Determines if the pool is in use. If so, it returns true and the state of
1572 * the pool as well as the name of the pool. Both strings are allocated and
1573 * must be freed by the caller.
1576 zpool_in_use(libzfs_handle_t *hdl, int fd, pool_state_t *state, char **namestr,
1582 uint64_t guid, vdev_guid;
1583 zpool_handle_t *zhp;
1584 nvlist_t *pool_config;
1585 uint64_t stateval, isspare;
1586 aux_cbdata_t cb = { 0 };
1591 if (zpool_read_label(fd, &config) != 0) {
1592 (void) no_memory(hdl);
1599 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
1601 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID,
1604 if (stateval != POOL_STATE_SPARE && stateval != POOL_STATE_L2CACHE) {
1605 verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1607 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1612 case POOL_STATE_EXPORTED:
1614 * A pool with an exported state may in fact be imported
1615 * read-only, so check the in-core state to see if it's
1616 * active and imported read-only. If it is, set
1617 * its state to active.
1619 if (pool_active(hdl, name, guid, &isactive) == 0 && isactive &&
1620 (zhp = zpool_open_canfail(hdl, name)) != NULL) {
1621 if (zpool_get_prop_int(zhp, ZPOOL_PROP_READONLY, NULL))
1622 stateval = POOL_STATE_ACTIVE;
1625 * All we needed the zpool handle for is the
1626 * readonly prop check.
1634 case POOL_STATE_ACTIVE:
1636 * For an active pool, we have to determine if it's really part
1637 * of a currently active pool (in which case the pool will exist
1638 * and the guid will be the same), or whether it's part of an
1639 * active pool that was disconnected without being explicitly
1642 if (pool_active(hdl, name, guid, &isactive) != 0) {
1643 nvlist_free(config);
1649 * Because the device may have been removed while
1650 * offlined, we only report it as active if the vdev is
1651 * still present in the config. Otherwise, pretend like
1654 if ((zhp = zpool_open_canfail(hdl, name)) != NULL &&
1655 (pool_config = zpool_get_config(zhp, NULL))
1659 verify(nvlist_lookup_nvlist(pool_config,
1660 ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1661 ret = find_guid(nvroot, vdev_guid);
1667 * If this is an active spare within another pool, we
1668 * treat it like an unused hot spare. This allows the
1669 * user to create a pool with a hot spare that currently
1670 * in use within another pool. Since we return B_TRUE,
1671 * libdiskmgt will continue to prevent generic consumers
1672 * from using the device.
1674 if (ret && nvlist_lookup_uint64(config,
1675 ZPOOL_CONFIG_IS_SPARE, &isspare) == 0 && isspare)
1676 stateval = POOL_STATE_SPARE;
1681 stateval = POOL_STATE_POTENTIALLY_ACTIVE;
1686 case POOL_STATE_SPARE:
1688 * For a hot spare, it can be either definitively in use, or
1689 * potentially active. To determine if it's in use, we iterate
1690 * over all pools in the system and search for one with a spare
1691 * with a matching guid.
1693 * Due to the shared nature of spares, we don't actually report
1694 * the potentially active case as in use. This means the user
1695 * can freely create pools on the hot spares of exported pools,
1696 * but to do otherwise makes the resulting code complicated, and
1697 * we end up having to deal with this case anyway.
1700 cb.cb_guid = vdev_guid;
1701 cb.cb_type = ZPOOL_CONFIG_SPARES;
1702 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1703 name = (char *)zpool_get_name(cb.cb_zhp);
1710 case POOL_STATE_L2CACHE:
1713 * Check if any pool is currently using this l2cache device.
1716 cb.cb_guid = vdev_guid;
1717 cb.cb_type = ZPOOL_CONFIG_L2CACHE;
1718 if (zpool_iter(hdl, find_aux, &cb) == 1) {
1719 name = (char *)zpool_get_name(cb.cb_zhp);
1732 if ((*namestr = zfs_strdup(hdl, name)) == NULL) {
1734 zpool_close(cb.cb_zhp);
1735 nvlist_free(config);
1738 *state = (pool_state_t)stateval;
1742 zpool_close(cb.cb_zhp);
1744 nvlist_free(config);